Avoidable Mortality Attributable to Anthropogenic Fine Particulate Matter (PM2.5) in Australia

Ambient fine particulate matter <2.5 µm (PM2.5) air pollution increases premature mortality globally. Some PM2.5 is natural, but anthropogenic PM2.5 is comparatively avoidable. We determined the impact of long-term exposures to the anthropogenic PM component on mortality in Australia. PM2.5-attributable deaths were calculated for all Australian Statistical Area 2 (SA2; n = 2310) regions. All-cause death rates from Australian mortality and population databases were combined with annual anthropogenic PM2.5 exposures for the years 2006–2016. Relative risk estimates were derived from the literature. Population-weighted average PM2.5 concentrations were estimated in each SA2 using a satellite and land use regression model for Australia. PM2.5-attributable mortality was calculated using a health-impact assessment methodology with life tables and all-cause death rates. The changes in life expectancy (LE) from birth, years of life lost (YLL), and economic cost of lost life years were calculated using the 2019 value of a statistical life. Nationally, long-term population-weighted average total and anthropogenic PM2.5 concentrations were 6.5 µg/m3 (min 1.2–max 14.2) and 3.2 µg/m3 (min 0–max 9.5), respectively. Annually, anthropogenic PM2.5-pollution is associated with 2616 (95% confidence intervals 1712, 3455) deaths, corresponding to a 0.2-year (95% CI 0.14, 0.28) reduction in LE for children aged 0–4 years, 38,962 (95%CI 25,391, 51,669) YLL and an average annual economic burden of $6.2 billion (95%CI $4.0 billion, $8.1 billion). We conclude that the anthropogenic PM2.5-related costs of mortality in Australia are higher than community standards should allow, and reductions in emissions are recommended to achieve avoidable mortality.

. Average PM2.5 in 2015 across the country and inset maps of the Sydney region and the small case study region in Western Sydney. The Western Sydney inset map shows the MBs.

S2. Estimated level of anthropogenic PM2.5
We estimated the anthropogenic PM2.5 as the difference between total PM2.5 and the 5 th percentile of concentrations observed in any MB per state/territory per year. Thus, we assumed that the natural background level would be different in each state/territory due to natural processes that may vary with the ecological types found in each state/territory, and the 5 th percentile value represents the level there would be without human emissions. Because the Australian Capital Territory (ACT) is a relatively smaller geographic region and the MB range was considered unreliable we used the broader neighbouring region of New South Wales (NSW) to derive the estimate of the non-anthropogenic concentration for that territory.

S3. Health outcomes and population data
We linked each state/territory's age-specific deaths data with the age-specific populations at SA2 by 5

S4. Concentration-response functions
The present concentration-response function (CRF) is recommended by the World Health Organization (WHO) Health risks of air pollution in Europe (HRAPIE) project for estimates of all cause long-term mortality due to PM2.5 (2).

S5. Life table method
We calculated life tables for each subpopulation as described by Chiang (1968) (3). First the age-specific death rates (Mx) are calculated by dividing the number of deaths in each year by the total number of individuals in each age group. The probability that an individual will die during the age interval (Qx) is conditional on the fraction of the age interval (0.5) lived by those who die in it and the width of the age interval (5 years). The reciprocal probability of survival (Px) until the end of the age interval is calculated by subtracting Qx from 1. These probabilities are then imposed on a hypothetical cohort (Ix) with a starting population of 100000 births and reduced numbers in all age groups according to Px, and the numbers of deaths in each age group are calculated by subtracting numbers of living people from those of the previous age band. The number of years lived (Lx) in each age interval can then be calculated by adding the number of surviving people in each age group to half the number who died within the 5 year interval. The cumulative number of years lived by the cohort population in the age interval and all subsequent intervals (Tx) is therefore equal to the number of years lived plus that of the previous age interval. Life expectancy (LE) at the beginning of each age interval is therefore equal to the cumulative number of years lived divided by 100000 (Tx/Ix).
We implemented the calculations using the life tables approach in the IOMLIFETR software by Broome et al.

S6. LE and years of life lost (YLL)
Using the series of life table matrices to model mortality, we compared LE under a business as usual (BAU) scenario with those of a reduced emissions scenario. Numbers of life years gained by the hypothetical emissions reductions were estimated by calculating the anthropogenic PM2.5-attributable numbers of deaths for each age group in the BAU scenario and multiplying these by the remaining LE for each age group and then summing these for all age groups. We report total numbers of life years lost relative to expected life years under the assumption of instantaneous cessation of PM2.5. As such, no lag structure has been employed in these calculations. Finally, we averaged LE for children (aged 0-4) in every SA2 across the entire country to indicate the attributable health burden. YYL across the entire country were calculated for each year of the study using the widely reported equation: where ANij and LEij are attributable numbers and the counterfactual life expectancy, respectively, for each age group i and SA2 j.

S7. Economic calculations
We used the willingness-to-pay value of a statistical life year (VSLY = $213000) as described by the Office of Best Practice Regulation in 2019 (4). VSLY estimates are based on how much people value reductions in the risk of mortality that cumulatively produce an additional life year in a population. We did not adjust this value according to underlying health status, age or LE, but we applied a social discount rate (SDR) of 3% annually. This SDR reflects economic depreciation rather than ageing and is a composite of a private time preference rate, a social time preference rate, the opportunity cost of capital, and a weighted cost of funds discount rate, as described elsewhere (5,6). Whereas VSLY does not vary with age, the value of a statistical life (VSL) does. To calculate age specific VSLs from the VSLY, we applied an annual discount of 3% sequentially for the number of expected remaining life years and summed all discounted values for each age group. As stated in the best practice guidelines (4), the 2019 VSL for a person with 40 years to live is $4.9 million. We calculated age-group specific VSLs by summing sequentially discounted VSLY values for the number of expected remaining life years (LE) of each age group. Age specific VSLs were then multiplied by PM2.5-attributable numbers (AN) of deaths.

S8. Sensitivity analysis using climate zones
One of the key assumptions made in our main analysis was the estimate of nonanthropogenic PM2.5 concentrations by taking the 5 th percentile level within states/territories to represent the lower possible levels that are observed within each geographic region. In alternative analysis for sensitivity assessment we instead used geographic regions that represent climatological boundaries from the Bureau of Meteorology, as shown in Figures S2 and S3. Despite the improved congruity with profiles of seasonal rainfall, which influence periods of increased natural PM2.5 from dust and bushfires, our estimates of health burden were similar with 2837 (95%CI 1858, 3746) PM2.5-attributable deaths compared to our main analysis estimate of 2616 (95%CI 1712, 3455) using the state/territory based non-anthropogenic PM2.5. Figure S2: Climate zones from Bureau of Meteorology rainfall levels. These climate maps are classified by seasonal rainfall and come from the National Climate Centre of the Bureau of Meteorology: Climate Zone Classification. http://www.bom.gov.au/jsp/ncc/climate_averages/climateclassifications/index.jsp. Summer was defined as November -April and winter was defined as May -October. Then the following rules concerning rainfall were followed: Summer dominant (sd), summer > winter AND ratio seasonal rainfall (greater/lesser) > 3 AND Annual total rainfall > 350 mm; Summer (s), summer > winter AND ratio (greater/lesser) < 3 and > 1.3 AND annual total rainfall > 350 mm; Winter dominant (wd), winter > summer AND ratio (greater/lesser) > 3 AND annual total rainfall > 350 mm; Winter (w), winter > summer AND ratio (greater/lesser) < 3 and > 1.3 AND annual total rainfall > 350 mm; Uniform (u), ratio < 1.3 AND annual total rainfall > 350 mm; and Arid (a), annual total rainfall < 350 mm.

S9. Sensitivity analysis using revised relative risk (RR) from recent meta-analysis
A new meta-analysis has recently been published (7) that found support for a RR of 1.08 (95%CI 1.06, 1.09) per 10 µg/m 3 increase in PM2.5 which is higher than the RR estimate we have used (2). In an additional sensitivity assessment, we used this new RR and found an increase to our estimated health burden as expected (3335 premature deaths, 95%CI 2534, 3728). However, the difference did not affect our conclusion based on our main analysis that the health burden is substantial.